PostgreSQL: Documentation: 16: 9.16. JSON Functions and Operators

SQL/JSON path expressions specify the items to be retrieved from the JSON data, similar to XPath expressions used for SQL access to XML. In PostgreSQL, path expressions are implemented as the jsonpath data type and can use any elements described in Section 8.14.7.

JSON query functions and operators pass the provided path expression to the path engine for evaluation. If the expression matches the queried JSON data, the corresponding JSON item, or set of items, is returned. Path expressions are written in the SQL/JSON path language and can include arithmetic expressions and functions.

A path expression consists of a sequence of elements allowed by the jsonpath data type. The path expression is normally evaluated from left to right, but you can use parentheses to change the order of operations. If the evaluation is successful, a sequence of JSON items is produced, and the evaluation result is returned to the JSON query function that completes the specified computation.

To refer to the JSON value being queried (the context item), use the $ variable in the path expression. It can be followed by one or more accessor operators, which go down the JSON structure level by level to retrieve sub-items of the context item. Each operator that follows deals with the result of the previous evaluation step.

For example, suppose you have some JSON data from a GPS tracker that you would like to parse, such as:

To retrieve the available track segments, you need to use the .key accessor operator to descend through surrounding JSON objects:

To retrieve the contents of an array, you typically use the [*] operator. For example, the following path will return the location coordinates for all the available track segments:

To return the coordinates of the first segment only, you can specify the corresponding subscript in the [] accessor operator. Recall that JSON array indexes are 0-relative:

The result of each path evaluation step can be processed by one or more jsonpath operators and methods listed in Section 9.16.2.2. Each method name must be preceded by a dot. For example, you can get the size of an array:

More examples of using jsonpath operators and methods within path expressions appear below in Section 9.16.2.2.

When defining a path, you can also use one or more filter expressions that work similarly to the WHERE clause in SQL. A filter expression begins with a question mark and provides a condition in parentheses:

Filter expressions must be written just after the path evaluation step to which they should apply. The result of that step is filtered to include only those items that satisfy the provided condition. SQL/JSON defines three-valued logic, so the condition can be true, false, or unknown. The unknown value plays the same role as SQL NULL and can be tested for with the is unknown predicate. Further path evaluation steps use only those items for which the filter expression returned true.

The functions and operators that can be used in filter expressions are listed in Table 9.51. Within a filter expression, the @ variable denotes the value being filtered (i.e., one result of the preceding path step). You can write accessor operators after @ to retrieve component items.

For example, suppose you would like to retrieve all heart rate values higher than 130. You can achieve this using the following expression:

To get the start times of segments with such values, you have to filter out irrelevant segments before returning the start times, so the filter expression is applied to the previous step, and the path used in the condition is different:

You can use several filter expressions in sequence, if required. For example, the following expression selects start times of all segments that contain locations with relevant coordinates and high heart rate values:

Using filter expressions at different nesting levels is also allowed. The following example first filters all segments by location, and then returns high heart rate values for these segments, if available:

You can also nest filter expressions within each other:

This expression returns the size of the track if it contains any segments with high heart rate values, or an empty sequence otherwise.

PostgreSQL's implementation of the SQL/JSON path language has the following deviations from the SQL/JSON standard:

A path expression can be a Boolean predicate, although the SQL/JSON standard allows predicates only in filters. This is necessary for implementation of the @@ operator. For example, the following jsonpath expression is valid in PostgreSQL:

There are minor differences in the interpretation of regular expression patterns used in like_regex filters, as described in Section 9.16.2.3.

When you query JSON data, the path expression may not match the actual JSON data structure. An attempt to access a non-existent member of an object or element of an array results in a structural error. SQL/JSON path expressions have two modes of handling structural errors:

lax (default) — the path engine implicitly adapts the queried data to the specified path. Any remaining structural errors are suppressed and converted to empty SQL/JSON sequences.

strict — if a structural error occurs, an error is raised.

The lax mode facilitates matching of a JSON document structure and path expression if the JSON data does not conform to the expected schema. If an operand does not match the requirements of a particular operation, it can be automatically wrapped as an SQL/JSON array or unwrapped by converting its elements into an SQL/JSON sequence before performing this operation. Besides, comparison operators automatically unwrap their operands in the lax mode, so you can compare SQL/JSON arrays out-of-the-box. An array of size 1 is considered equal to its sole element. Automatic unwrapping is not performed only when:

The path expression contains type() or size() methods that return the type and the number of elements in the array, respectively.

The queried JSON data contain nested arrays. In this case, only the outermost array is unwrapped, while all the inner arrays remain unchanged. Thus, implicit unwrapping can only go one level down within each path evaluation step.

For example, when querying the GPS data listed above, you can abstract from the fact that it stores an array of segments when using the lax mode:

In the strict mode, the specified path must exactly match the structure of the queried JSON document to return an SQL/JSON item, so using this path expression will cause an error. To get the same result as in the lax mode, you have to explicitly unwrap the segments array:

The .** accessor can lead to surprising results when using the lax mode. For instance, the following query selects every HR value twice:

This happens because the .** accessor selects both the segments array and each of its elements, while the .HR accessor automatically unwraps arrays when using the lax mode. To avoid surprising results, we recommend using the .** accessor only in the strict mode. The following query selects each HR value just once:

Table 9.50 shows the operators and methods available in jsonpath. Note that while the unary operators and methods can be applied to multiple values resulting from a preceding path step, the binary operators (addition etc.) can only be applied to single values.

Table 9.50. jsonpath Operators and Methods

Operator/Method

Description

Example(s)

number + number → number

Addition

jsonb_path_query('[2]', '$[0] + 3') → 5

+ number → number

Unary plus (no operation); unlike addition, this can iterate over multiple values

jsonb_path_query_array('{"x": [2,3,4]}', '+ $.x') → [2, 3, 4]

number - number → number

Subtraction

jsonb_path_query('[2]', '7 - $[0]') → 5

- number → number

Negation; unlike subtraction, this can iterate over multiple values

jsonb_path_query_array('{"x": [2,3,4]}', '- $.x') → [-2, -3, -4]

number * number → number

Multiplication

jsonb_path_query('[4]', '2 * $[0]') → 8

number / number → number

Division

jsonb_path_query('[8.5]', '$[0] / 2') → 4.2500000000000000

number % number → number

Modulo (remainder)

jsonb_path_query('[32]', '$[0] % 10') → 2

value . type() → string

Type of the JSON item (see json_typeof)

jsonb_path_query_array('[1, "2", {}]', '$[*].type()') → ["number", "string", "object"]

value . size() → number

Size of the JSON item (number of array elements, or 1 if not an array)

jsonb_path_query('{"m": [11, 15]}', '$.m.size()') → 2

value . double() → number

Approximate floating-point number converted from a JSON number or string

jsonb_path_query('{"len": "1.9"}', '$.len.double() * 2') → 3.8

number . ceiling() → number

Nearest integer greater than or equal to the given number

jsonb_path_query('{"h": 1.3}', '$.h.ceiling()') → 2

number . floor() → number

Nearest integer less than or equal to the given number

jsonb_path_query('{"h": 1.7}', '$.h.floor()') → 1

number . abs() → number

Absolute value of the given number

jsonb_path_query('{"z": -0.3}', '$.z.abs()') → 0.3

string . datetime() → datetime_type (see note)

Date/time value converted from a string

jsonb_path_query('["2015-8-1", "2015-08-12"]', '$[*] ? (@.datetime() < "2015-08-2".datetime())') → "2015-8-1"

string . datetime(template) → datetime_type (see note)

Date/time value converted from a string using the specified to_timestamp template

jsonb_path_query_array('["12:30", "18:40"]', '$[*].datetime("HH24:MI")') → ["12:30:00", "18:40:00"]

object . keyvalue() → array

The object's key-value pairs, represented as an array of objects containing three fields: "key", "value", and "id"; "id" is a unique identifier of the object the key-value pair belongs to

jsonb_path_query_array('{"x": "20", "y": 32}', '$.keyvalue()') → [{"id": 0, "key": "x", "value": "20"}, {"id": 0, "key": "y", "value": 32}]

The result type of the datetime() and datetime(template) methods can be date, timetz, time, timestamptz, or timestamp. Both methods determine their result type dynamically.

The datetime() method sequentially tries to match its input string to the ISO formats for date, timetz, time, timestamptz, and timestamp. It stops on the first matching format and emits the corresponding data type.

The datetime(template) method determines the result type according to the fields used in the provided template string.

The datetime() and datetime(template) methods use the same parsing rules as the to_timestamp SQL function does (see Section 9.8), with three exceptions. First, these methods don't allow unmatched template patterns. Second, only the following separators are allowed in the template string: minus sign, period, solidus (slash), comma, apostrophe, semicolon, colon and space. Third, separators in the template string must exactly match the input string.

If different date/time types need to be compared, an implicit cast is applied. A date value can be cast to timestamp or timestamptz, timestamp can be cast to timestamptz, and time to timetz. However, all but the first of these conversions depend on the current TimeZone setting, and thus can only be performed within timezone-aware jsonpath functions.

Table 9.51 shows the available filter expression elements.

Table 9.51. jsonpath Filter Expression Elements

Predicate/Value

Description

Example(s)

value == value → boolean

Equality comparison (this, and the other comparison operators, work on all JSON scalar values)

jsonb_path_query_array('[1, "a", 1, 3]', '$[*] ? (@ == 1)') → [1, 1]

jsonb_path_query_array('[1, "a", 1, 3]', '$[*] ? (@ == "a")') → ["a"]

value != value → boolean

value value → boolean

Non-equality comparison

jsonb_path_query_array('[1, 2, 1, 3]', '$[*] ? (@ != 1)') → [2, 3]

jsonb_path_query_array('["a", "b", "c"]', '$[*] ? (@ "b")') → ["a", "c"]

value = value → boolean

Greater-than-or-equal-to comparison

jsonb_path_query_array('[1, 2, 3]', '$[*] ? (@ >= 2)') → [2, 3]

true → boolean

JSON constant true

jsonb_path_query('[{"name": "John", "parent": false}, {"name": "Chris", "parent": true}]', '$[*] ? (@.parent == true)') → {"name": "Chris", "parent": true}

false → boolean

JSON constant false

jsonb_path_query('[{"name": "John", "parent": false}, {"name": "Chris", "parent": true}]', '$[*] ? (@.parent == false)') → {"name": "John", "parent": false}

null → value

JSON constant null (note that, unlike in SQL, comparison to null works normally)

jsonb_path_query('[{"name": "Mary", "job": null}, {"name": "Michael", "job": "driver"}]', '$[*] ? (@.job == null) .name') → "Mary"

boolean && boolean → boolean

Boolean AND

jsonb_path_query('[1, 3, 7]', '$[*] ? (@ > 1 && @ < 5)') → 3

boolean || boolean → boolean

Boolean OR

jsonb_path_query('[1, 3, 7]', '$[*] ? (@ < 1 || @ > 5)') → 7

! boolean → boolean

Boolean NOT

jsonb_path_query('[1, 3, 7]', '$[*] ? (!(@ < 5))') → 7

boolean is unknown → boolean

Tests whether a Boolean condition is unknown.

jsonb_path_query('[-1, 2, 7, "foo"]', '$[*] ? ((@ > 0) is unknown)') → "foo"

string like_regex string [ flag string ] → boolean

Tests whether the first operand matches the regular expression given by the second operand, optionally with modifications described by a string of flag characters (see Section 9.16.2.3).

jsonb_path_query_array('["abc", "abd", "aBdC", "abdacb", "babc"]', '$[*] ? (@ like_regex "^ab.*c")') → ["abc", "abdacb"]

jsonb_path_query_array('["abc", "abd", "aBdC", "abdacb", "babc"]', '$[*] ? (@ like_regex "^ab.*c" flag "i")') → ["abc", "aBdC", "abdacb"]

string starts with string → boolean

Tests whether the second operand is an initial substring of the first operand.

jsonb_path_query('["John Smith", "Mary Stone", "Bob Johnson"]', '$[*] ? (@ starts with "John")') → "John Smith"

exists ( path_expression ) → boolean

Tests whether a path expression matches at least one SQL/JSON item. Returns unknown if the path expression would result in an error; the second example uses this to avoid a no-such-key error in strict mode.

jsonb_path_query('{"x": [1, 2], "y": [2, 4]}', 'strict $.* ? (exists (@ ? (@[*] > 2)))') → [2, 4]

jsonb_path_query_array('{"value": 41}', 'strict $ ? (exists (@.name)) .name') → []

SQL/JSON path expressions allow matching text to a regular expression with the like_regex filter. For example, the following SQL/JSON path query would case-insensitively match all strings in an array that start with an English vowel:

The optional flag string may include one or more of the characters i for case-insensitive match, m to allow ^ and $ to match at newlines, s to allow . to match a newline, and q to quote the whole pattern (reducing the behavior to a simple substring match).

The SQL/JSON standard borrows its definition for regular expressions from the LIKE_REGEX operator, which in turn uses the XQuery standard. PostgreSQL does not currently support the LIKE_REGEX operator. Therefore, the like_regex filter is implemented using the POSIX regular expression engine described in Section 9.7.3. This leads to various minor discrepancies from standard SQL/JSON behavior, which are cataloged in Section 9.7.3.8. Note, however, that the flag-letter incompatibilities described there do not apply to SQL/JSON, as it translates the XQuery flag letters to match what the POSIX engine expects.

Keep in mind that the pattern argument of like_regex is a JSON path string literal, written according to the rules given in Section 8.14.7. This means in particular that any backslashes you want to use in the regular expression must be doubled. For example, to match string values of the root document that contain only digits: